# ownership.^{14,15 }This control group should capture the broader developments in

industry and the economy (e.g. technical progress), so that any incremental

performance improvements of the privatised companies can be considered firm-

specific and associated with the privatisation process.

# Because the control group data is only available for the years post 1988, we first run

the regression model on the full sample of initial SIPs, but without the control group

variable included in the specification (see Table 3). We then define a sub-sample of

those initial SIPs that took place after 1988, for which control group data is available

(see Table 4). Diagnostic model tests indicate that the errors are non-spherical, i.e.

subject to both serial autocorrelation and heteroskedasticity.^{16 }Whilst a number of

different estimation procedures have been considered, the significance of the unit

effects has been important in the choice of a fixed-effects model with cluster-robust

error terms.^{17 }The results presented in this paper are based on the dataset being

adjusted for outliers at the 5% level.

# For the full sample, the coefficient of the ‘Post’ variable has the predicted sign for all

performance measurements^{18}, i.e. there is a discrete step-up in performance after the

handover of property rights to private investors, but this is only significant for the

increase in profitability and the reduction of employment intensity.

14 The control group is based on the "PIW Top 50 Oil Companies" rankings (PIW 1988-2007), which has the most comprehensive coverage available of both public and private oil and gas companies. It contains information on output, employees, revenues, net income and assets, but not on capital expenditure, production costs, financial leverage or dividends, as much of this is unavailable for fully state-owned NOCs. The control group consists of the 21 (public and private) companies that did not experience any changes to their ownership structure and that have featured in all of the 20 annual rankings. We take the median annual performance to be the performance of the control group.

15 We use both variables in order to distinguish the truly exogenous impact of oil prices from other aspects of performance changes in the control group, which might be linked to management decisions.

16 The autocorrelation test is based on Wooldridge (2002, p.282-3) and heteroskedasticity is tested via a likelihood-ratio test. Unit roots have been tested following Maddala and Wu (1999).

17 Beck and Katz (1995) show that a commonly used variety of feasible generalized least squares, as recommended by Parks (1967) and Kmenta (1986), produces unduly optimistic standard errors unless T >>N. But the alternative suggested by Beck and Katz, OLS with panel-corrected standard errors, does not perform well in the presence of unit effects (Adolph et al. 2005; Wilson and Butler 2007).

18 For capital expenditure, the sign of the POST variable is negative, but this is overcompensated by the positive change in the time trend.

16